RLAIF and Endogenous Selection in Agents

Why RLAIF and endogenous LLM judges drift under semantic closure, how synthetic data drives model collapse, and how consequence-based selection works.